Low Computational Resources Research Articles

Learning the Relation Between Interested Objects and Aesthetic Region for Image Cropping

As one of the fundamental techniques for image editing, image cropping discards irrelevant contents and remains the pleasing portions of the image to enhance the overall composition and achieve better visual/aesthetic perception. In this paper, we primarily focus on improving the efficiency of automatic image cropping, and on further exploring its potential in public datasets with high accuracy. From this perspective, we propose a deep learning based framework to learn the objects composition from photos with high aesthetic qualities, where an interested object region is detected through a convolutional neural network (CNN) based on the saliency map. The features of the detected interested objects are then fed into a regression network to obtain the final cropping result. Unlike the conventional methods that multiple candidates are proposed and evaluated iteratively, only a single interested object region is produced in our model, which is mapped to the final output directly. Thus, low computational resources are required for the proposed approach. The experimental results on the public datasets show that as a weakly supervised method, the proposed network outperforms the other weakly supervised methods on FLMS and FCD datasets and achieves comparable results to the existing methods on CUHK dataset. Furthermore, the proposed method is more efficient than these methods, where the processing speed is as fast as 20 ms per image.

A New Model to Distinguish Railhead Defects Based on Set-Membership Type-2 Fuzzy Logic System

This paper focuses on the new model for the classification of railhead defects, through images acquired by a rail inspection vehicle. In this regard, we discuss the use of set-membership concept, derived from the adaptive filter theory, into the training procedure of an upper and lower singleton type-2 fuzzy logic system, aiming to reduce computational complexity and to increase the convergence speed. The performance is based on the data set composed of images provided by a Brazilian railway company, which covers the four possible railhead defects (cracking, flaking, head-check and spalling) and the normal condition of the railhead. Additionally, we apply different levels of additive white Gaussian noise in the images in order to challenge the proposed model. Finally, we discuss performance analysis in terms of convergence speed, computational complexity reduction, and classification ratio. The reported results show that the proposal achieved improved convergence speed, slightly higher classification ratio and remarkable computation complexity reduction when we limit the number of epochs for training, which may be required under real-time constraint or low computational resource availability.

Modelling of a Three-Body Hinge-Barge Wave Energy Device Using System Identification Techniques

In order to increase the prevalence of wave energy converters (WECs), they must provide energy at competitive prices, especially when compared with other renewable energy sources. Thus, it is imperative to develop control system technologies that are able to maximize energy extraction from waves, such that the delivered energy cost is reduced. An important part of a model-based controller is the model that it uses. System identification techniques (SITs) provide methodologies to get accurate dynamic models from input-output data. However, even though these techniques are well developed in other application areas, they are seldom used in the context of WECs. This paper proposes several strategies based on SIT to get a linear time-invariant model for a three-body hinge-barge wave energy device using experimental data. The main advantage of the model obtained with this methodology, against other methods such as linear potential theory, is that this model remains valid even for relatively large waves and WEC displacements. Other advantages of this model are its simplicity and the low computational resources that it needs. Numerical simulations are carried out to show the validation of the obtained model against recorded experimental data.

Denial-of-Service Attacks on Wireless Sensor Network and Defense Techniques

The basic characteristic of a wireless sensor network is to sense the environment and to gather different types of data. It is used in various vital sectors, such as military, health-care, traffic monitoring, and so on. This provides many helpful data about these fields, which is used in implementing new technologies for them. However, there are numerous challenges and risks, regarding the security vulnerability of a wireless sensor network. This is because the resources of a wireless sensor network are very low (for example, low power, low computational resources, low storage, etc.). These constraints make them susceptible to different kinds of security attacks, especially the denial of service attacks. In this survey, we have outlined different kinds of DoS attacks in different layers of the hybrid layering model. We have also presented different defense mechanisms to protect/mitigate these attacks at different layers. Layer by layer analysis makes it easy to distinguish among different types of attacks and their defense techniques, along with the characteristics of the layers. Analysis of the critical issues of these attacks and defenses are finally presented with possible measures. This survey work can be very beneficial for the researchers and engineers in designing WSN topology with ensured security.

Role of Machine Learning Algorithms Intrusion Detection in WSNs: A Survey

Machine Learning is capable of providing real-time solutions that maximize the utilization of resources in the network thereby increasing the lifetime of the network. It is able to process automatically without being externally programmed thus making the process more easy, efficient, cost-effective, and reliable. ML algorithms can handle complex data more quickly and accurately. Machine Learning is used to enhance the ability of the Wireless Sensor Network environment. Wireless Sensor Networks (WSN) is a combination of several networks and it is decentralized and distributed in nature. WSN consists of sensor nodes and sinks nodes which have a property of self-organizing and self-healing. WSN is used in other applications, such as biodiversity and ecosystem protection, surveillance, climate change tracking, and other military applications.Now-a-days, a huge development is seen in WSNs due to the advancement of electronics and wireless communication technologies, several drawbacks like low computational capacity, small memory, and limited energy resources infrastructure needs physical vulnerability to require source measures where privacy plays a key role.WSN is used to monitor the dynamic environments and to adapt to such situation sensor networks need Machine Learning techniques to avoid unnecessary redesign. Machine learning techniques survey for WSNs provide a wide range of applications in which security is given top priority. To secure data from attackers the WSNs system should be able to delete the instruction if any hackers/attackers are trying to steal data.

Low-resource automatic cartoon image creation from limited samples

In this work, a framework that can automatically create cartoon images with low computation resources and small training datasets is proposed. The proposed system performs region segmentation and learns a region relationship tree from each learning image. The segmented regions are clustered automatically with an enhanced clustering mechanism with no prior knowledge of number of clusters. According to the topology represented by region relationship tree and clustering results, the regions are reassembled to create new images. A swarm intelligence optimization procedure is designed to coordinate the regions to the optimized sizes and positions in the created image. Rigid deformation using moving least squares is performed on the regions to generate more variety for created images. Compared with methods based on Generative Adversarial Networks, the proposed framework can create better images with limited computation resources and a very small amount of training samples.

Neural Network-Based Voting System with High Capacity and Low Computation for Intrusion Detection in SIEM/IDS Systems

Integrating intelligence into intrusion detection tools has received much attention in the last years. The goal is to improve the detection capability within SIEM and IDS systems in order to cope with the increasing number of attacks using sophisticated and complex methods to infiltrate systems. Current SIEM and IDS systems have many processes involved, which work together to collect, analyze, detect, and send notification of failures in real time. Event normalization, for example, requires significant processing power to handle network events. So, adding heavy deep learning models will invoke additional resources for the SIEM or IDS tool. This paper presents a majority system based on reliability approach that combines simple feedforward neural networks, as weak learners, and produces high detection capability with low computation resources. The experimental results show that the model is very suitable for modeling a classification model with high accuracy and that its performance is superior to that of complex resource-intensive deep learning models.

Quasistasic modeling of edge fields in planar resonators

This paper discusses quasistatic waveguide model for planar resonator placed in multilayer dielectic media, which do not use effective parameters such as effective width of resonator and effective dielectric constant. It allows to use waveguide model in new approach. Edge field is modeled by inclusion of LC-lumped element circuit where values of inductances and capacitances we determined from the solving of electrostatic Laplace equation by the method of moment using the three-dimensional Green function for multilayer dielectric media. A computational experiment showed that the proposed model is effective in modeling planar resonators of complex shape and requires low computational resources.

A fast and scalable architecture to run convolutional neural networks in low density FPGAs

Deep learning and, in particular, convolutional neural networks (CNN) achieve very good results on several computer vision applications like security and surveillance, where image and video analysis are required. These networks are quite demanding in terms of computation and memory and therefore are usually implemented in high-performance computing platforms or devices. Running CNNs in embedded platforms or devices with low computational and memory resources requires a careful optimization of system architectures and algorithms to obtain very efficient designs. In this context, Field Programmable Gate Arrays (FPGA) can achieve this efficiency since the programmable hardware fabric can be tailored for each specific network. In this paper, a very efficient configurable architecture for CNN inference targeting any density FPGAs is described. The architecture considers fixed-point arithmetic and image batch to reduce computational, memory and memory bandwidth requirements without compromising network accuracy. The developed architecture supports the execution of large CNNs in any FPGA devices including those with small on-chip memory size and logic resources. With the proposed architecture, it is possible to infer an image in AlexNet in 4.3 ms in a ZYNQ7020 and 1.2 ms in a ZYNQ7045.

Secure multiparty quantum computation based on Lagrange unitary operator

As an important subtopic of classical cryptography, secure multiparty quantum computation allows multiple parties to jointly compute their private inputs without revealing them. Most existing secure multiparty computation protocols have the shortcomings of low computational efficiency and high resource consumption. To remedy these shortcomings, we propose a secure multiparty quantum computation protocol by using the Lagrange unitary operator and the Shamir (t, n) threshold secret sharing, in which the server generates all secret shares and distributes each secret share to the corresponding participant, in addition, he prepares a particle and sends it to the first participant. The first participant performs the Lagrange unitary operation on the received particle, and then sends the transformed particle to the next participant. Until the last participant’s computation task is completed, the transformed particle is sent back to the server. The server performs Lagrange unitary operation on the received particle by using a secret message, and then measures the transformed particle to obtain the sum of the calculations of multiple participants. Security analysis shows that the proposed protocol can resist intercept-measurement attack, intercept-resend attack, entanglement-swapping attack, entanglement-measurement attack and collusion attack. Performance comparison shows that it has higher computation efficiency and lower resource consumption than other similar protocols.

Learning to Play the Chess Variant Crazyhouse Above World Champion Level With Deep Neural Networks and Human Data.

Deep neural networks have been successfully applied in learning the board games Go, chess, and shogi without prior knowledge by making use of reinforcement learning. Although starting from zero knowledge has been shown to yield impressive results, it is associated with high computationally costs especially for complex games. With this paper, we present CrazyAra which is a neural network based engine solely trained in supervised manner for the chess variant crazyhouse. Crazyhouse is a game with a higher branching factor than chess and there is only limited data of lower quality available compared to AlphaGo. Therefore, we focus on improving efficiency in multiple aspects while relying on low computational resources. These improvements include modifications in the neural network design and training configuration, the introduction of a data normalization step and a more sample efficient Monte-Carlo tree search which has a lower chance to blunder. After training on 569537 human games for 1.5 days we achieve a move prediction accuracy of 60.4%. During development, versions of CrazyAra played professional human players. Most notably, CrazyAra achieved a four to one win over 2017 crazyhouse world champion Justin Tan (aka LM Jann Lee) who is more than 400 Elo higher rated compared to the average player in our training set. Furthermore, we test the playing strength of CrazyAra on CPU against all participants of the second Crazyhouse Computer Championships 2017, winning against twelve of the thirteen participants. Finally, for CrazyAraFish we continue training our model on generated engine games. In 10 long-time control matches playing Stockfish 10, CrazyAraFish wins three games and draws one out of 10 matches.

Fast head detection in arbitrary poses using depth information

Purpose This study aims to develop a real-time algorithm, which can detect people even in arbitrary poses. To cover poor and changing light conditions, it does not rely on color information. The developed method is expected to run on computers with low computational resources so that it can be deployed on autonomous mobile robots. Design/methodology/approach The method is designed to have a people detection pipeline with a series of operations. Efficient point cloud processing steps with a novel head extraction operation provide possible head clusters in the scene. Classification of these clusters using support vector machines results in high speed and robust people detector. Findings The method is implemented on an autonomous mobile robot and results show that it can detect people with a frame rate of 28 Hz and equal error rate of 92 per cent. Also, in various non-standard poses, the detector is still able to classify people effectively. Research limitations/implications The main limitation would be for point clouds similar to head shape causing false positives and disruptive accessories (like large hats) causing false negatives. Still, these can be overcome with sufficient training samples. Practical implications The method can be used in industrial and social mobile applications because of its robustness, low resource needs and low power consumption. Originality/value The paper introduces a novel and efficient technique to detect people in arbitrary poses, with poor light conditions and low computational resources. Solving all these problems in a single and lightweight method makes the study fulfill an important need for collaborative and autonomous mobile robots.

Optimal design methodology for sizing a fuel cell/battery hybrid power source

This paper presents an optimal design methodology enabling to exhibit the best parameters of a complex energy system combing several components and their related control parts. It is based on a particle swarm optimization technique for component sizing, combined with optimal control to consider energy management constraints. This approximate resolution is valuable since it allows to achieve a robust and effective optimal design using low computational resources: it enables to tackle large search spaces in engineering time constraints. The selected use case is a fuel cell/battery hybrid power source based on a power-split parallel architecture. Its performance index is defined as the fuel consumption. Regarding this objective, the drivetrain components size and the control parameters values are both strongly coupled and physically constrained. In this context, the methodology makes a tradeoff between component sizing and energy saving. Simulation results show the relevance and robustness of this approach regarding different driving cycles and operating conditions. It validates the replicability of this method to other optimization problems in the field of energy optimization. A comprehensive review of the simulation tests highlights the present limits of this optimization and provides new perspectives for future works.

Sound model of an orchestral kettledrum considering viscoelastic effects

The modeling of the mechanical systems that describe musical instruments has long been a topic of interest for acoustic physicists. The orchestral kettledrum, distinguished by a broad spectrum of sounds, has been the subject of multiple investigations related to the goal of generating a realistic digital sound synthesis. In the present work, we apply the Green function method to estimate modal frequencies considering an air-loaded viscoelastic membrane. We propose a method that includes viscoelasticity to accurately predict the sound spectrum of the modeled kettledrum. Results are compared to real sound recordings of an orchestral kettledrum obtained in controlled conditions. The calculated modal frequencies are found to coincide well with the real values with an absolute mean error of 1.25±0.76 Hz and 1.87±1.83 Hz for the A2 and B2♭ tuned drumheads respectively. The spectral envelope of the synthesized sound spectrum coincides well with the Fourier transform of the real sound. The viscoelastic term was found to generally reduce the amplitude of the sound spectrum and in certain cases, better approximate the modal frequencies and decay times. The modal synthesis method used here is numerically lightweight and can be adapted to be used in real-time applications with low computational resources. To reproduce our experiment, the recorded kettledrum sounds and Python source code of the model are freely available.

Time-domain diffuse optical tomography utilizing truncated Fourier series approximation.

Diffuse optical tomography (DOT) uses near infrared light for in vivo imaging of spatially varying optical parameters in biological tissues. It is known that time-resolved measurements provide the richest information on soft tissues, among other measurement types in DOT such as steady-state and intensity-modulated measurements. Therefore, several integral-transform-based moments of the time-resolved DOT measurements have been considered to estimate spatially distributed optical parameters. However, the use of such moments can result in low-contrast images and cross-talks between the reconstructed optical parameters, limiting their accuracy. In this work, we propose to utilize a truncated Fourier series approximation in time-resolved DOT. Using this approximation, we obtained optical parameter estimates with accuracy comparable to using whole time-resolved data that uses low computational time and resources. The truncated Fourier series approximation based estimates also displayed good contrast and minimal parameter cross-talk, and the estimates further improved in accuracy when multiple Fourier frequencies were used.

Embedding Gesture Prior to Joint Shape Optimization Based Real-Time 3D Hand Tracking

In this paper, we present a novel approach for 3D hand tracking in real-time from a set of depth images. In each frame, our approach initializes hand pose with learning and then jointly optimizes the hand pose and shape. For pose initialization, we propose a gesture classification and root location network (GCRL), which can capture the meaningful topological structure of the hand to estimate the gesture and root location of the hand. With the per-frame initialization, our approach can rapidly recover from tracking failures. For optimization, unlike most existing methods that have been using a fixed-size hand model or manual calibration, we propose a hand gesture-guided optimization strategy to estimate pose and shape iteratively, which makes the tracking results more accuracy. Experiments on three challenging datasets show that our proposed approach achieves similar accuracy as state-of-the-art approaches, while runs on a low computational resource (without GPU).

Low computing resource consumption fully digital foreground calibration technique for time-interleaved analog-to-digital converter

A time-interleaved analog-to-digital converter (TI-ADC) uses several sub-analog-to-digital converters (sub-ADCs) to achieve a high sampling rate. Its applications include communication systems, oscilloscopes, healthcare instruments, etc. However, the presence of sub-ADC channel mismatches such as offset, gain and sample-time mismatches can significantly degrade the performance of TI-ADCs. This paper proposes a fully digital foreground calibration technique for the TI-ADC, including mismatch correction and estimation blocks. Unlike existing techniques, the proposed technique requires low computational resources. As the resource for complex computing in a field-programmable gate array (FPGA) is mainly the digital signal processor (DSP), the mismatch correction block unifies each computing unit in the FPGA into a multiply adder to achieve low DSP consumption. To further reduce the use of DSP resources, the selection of an optimal number of taps of the finite impulse response filter used for mismatch correction is discussed. The optimal-tap filter that uses the lowest amount of DSPs while satisfying the required signal-to-noise and distortion ratio flat area bandwidth is presented. A real-time hardware correction block was implemented for an 8 Gs/s TI-ADC system with two sub-ADC channels, and the results with this hardware verify the low computing resource consumption feature of the proposed calibration technique.

Resource Management of Maritime Edge Nodes for Collected Data Feedback

With the development of marine economy, more marine applications emerge leading to more deployed sensors and more generated data. However, current marine Internet only provides low transmission rate constrained by limited communication resources and bad transmission environment. How to efficiently use these resources and guarantee quality-of-service (QoS) of applications need to be addressed. This paper proposes an online optimization resource management algorithm to improve the communication resource efficiency with guaranteed QoSs for different applications. This algorithm needs neither of the resource cost function nor QoS constraint function at the resource scheduling node. Furthermore, the gradient information, which is usually needed in learning strategies, is not required either. Instead, this algorithm performs resource management of both computation and communication resources in each time slot only based on the observation of the last time slot. With slot-by-slot resource allocation, the communication cost can be minimized to be suitable for maritime scenarios. In the meantime, a long-term delay constraint can be satisfied. Results show that the proposed algorithm achieves the goal of reducing communication cost while guaranteeing the delay-constraint of different applications. Although more computation resources will improve its performance, this algorithm still can obtain the minimized cost given a low computation resource.

Siamese network for real-time tracking with action-selection

Considering that most deep learning based trackers capture accurate locations for targets at the expense of consuming much time in training phrase, in this paper we present a new powerful tracker using the Siamese network which can be implemented with low computation resource. Our proposed tracker can track targets accurately by a fine-tuned model which is convenient to train. During the tracking, we apply a new sampling method that is independent of training called action-selection to conduct selective and flexible sampling step by step with a variable stride, by which we can get bounding boxes with varied aspect radio. By verifying its performance on online tracking benchmarks, it turns out that our tracker achieves higher accuracy than most traditional trackers. In addition, our tracker operates at frame-rates beyond real-time.

Cryptographic Algorithms in Wearable Communications: An Empirical Analysis

In this letter, we assess the practical impact of lightweight block and stream cipher algorithms on power consumption and hardware resources for wearable devices that own low computational resources. Differently from the literature, we present an empirical and hardware-driven evaluation of the most representative encryption algorithms with regard to the requirements of wearable networks. We design and implement a cryptography library useful for wearable devices. Results confirm a strong correlation between the amount of logic/arithmetic operations, assembly instructions and power consumption for the two evaluated platforms, and they highlight the need to design encryption algorithms for wearable devices with high energy consumption efficiency, but strong security level similar to AES.